你在 D:\Research\DoneBench 做只读资源/相关工作侦察，不修改文件。请阅读 reports/agent_handoff.md 和 pa…
- Stable ID:
codex-user-prompt:a8f28f42af9c6262
- Source kind:
codex-session-user
- Category:
research-workflow
- Timestamp:
2026-05-09T00:07:24.192Z
- Semantic hash:
a8f28f42af9c62624321096d60b55247a1e23a67548e020689126bf7cf8913eb
- Public handling: selected full-text prompt with secret filtering.
Prompt Text
你在 D:\Research\DoneBench 做只读资源/相关工作侦察，不修改文件。请阅读 reports/agent_handoff.md 和 paper/ 目录，然后整理与 DoneBench 最相关的顶会/高影响 benchmark/evaluation 论文线索：WebArena, OSWorld, WorkArena/WorkArena++, tau-bench, AgentBench/AgentEval, SWE-bench 或类似工作。重点回答：这些论文是否要求人工双标作为必备 gate？它们如何做 dataset/task validation、human/expert validation、LLM-as-judge 或 automatic grading？哪些做法可借鉴到 DoneBench。请列出可核查的论文/项目名称和建议引用位置，不要修改文件。
Reuse Notes
- EXTRACTED: This is a selected Codex prompt or automation prompt from the local Codex corpus.
- INFERRED: Future agents can reuse its structure, constraints, and acceptance criteria when creating similar Codex workflows.